Hy4 Flash beats Hy4 in our tests for about 1/24 of the cost
Published 2026-11-17 · Data captured 17 Nov 2026, 20:22 UTC
With reasoning on, Tencent's Hy4 Flash scored 7.27/10, 1.44 points above the best Hy4 setting (low, 5.84). Its full run cost $0.098 against $2.365, about 1/24, and it answered 3.6 times faster (49.0 s vs 177.3 s). Without reasoning the two are close (5.94 vs 5.70), and Flash is 10 times cheaper. For a little less money, GPT-6 Luna high still scores higher (7.72).
See current Hy4 flash results →
- Tested through OpenRouter. Prices per million tokens: Hy4 Flash $0.035 input and $0.1405 output on the Tencent endpoint, which OpenRouter lists at a 50% discount; Hy4 $0.417 and $1.2505.
- Costs are recorded spend for one full run (23 tests, 3 repeats each). Failed API calls stay in the results and count as unsolved attempts.
Key results
The smaller model wins. Hy4 Flash with reasoning solved 50 of 69 attempts. Hy4 low solved 42 and Hy4 high 39.
Much cheaper and faster. $0.098 per run against $2.365 for Hy4 low and $3.055 for Hy4 high. Average answers took 49.0 s, against 177.3 s and 219.6 s.
Not only API errors. Hy4 low and high each lost 4 attempts to API failures. Even if all of them had been correct, Hy4 low would have solved 46, still fewer than Flash.
Without reasoning they are close. Flash scored 5.94 and Hy4 5.70, with similar response times (6.1 s vs 5.9 s). Flash cost $0.012 against $0.114.
Still a mid-table model. Hy4 Flash ranks #218 on the leaderboard. GPT-6 Luna high (#186) scores 0.44 points more for 75% of the cost and answers 2.4 times faster.
| Model | Effort | Rank | Score | Solved tests | Correct attempts | Run cost | Cents per solved test | Avg response |
|---|---|---|---|---|---|---|---|---|
| Hy4 flash | medium | #218 | 7.27 | 15/23 | 50/69 | $0.098 | 0.654 | 49.0 s |
| Hy4 flash | off | #346 | 5.94 | 9/23 | 29/69 | $0.012 | 0.128 | 6.1 s |
| Hy4 | low | #354 | 5.84 | 11/23 | 42/69 | $2.365 | 21.492 | 177.3 s |
| Hy4 | high | #357 | 5.81 | 10/23 | 39/69 | $3.055 | 30.547 | 219.6 s |
| Hy4 | off | #367 | 5.70 | 9/23 | 28/69 | $0.114 | 1.263 | 5.9 s |
| GPT-6 Luna | high | #186 | 7.72 | 15/23 | 51/69 | $0.074 | 0.492 | 20.4 s |
Score, cost and speed
Hy4's reasoning settings are the most expensive and slowest rows here, and they score below Flash with reasoning.
Score (out of 10, higher is better)
Full-run cost in USD (lower is better)
Average response time in seconds (lower is better)
Strengths and weak points
Flash with reasoning leads Hy4 low in coding (7.39 vs 5.74), combined tasks (6.50 vs 3.00), data parsing (9.95 vs 7.13), domain-specific questions (5.89 vs 2.93) and tool calling (10 vs 7.04).
Hy4 low is better at puzzles (7.67 vs 5.91). Both score about 4.7 in agentic tasks and 3.00 in trivia.
Hy4 flash medium
Hy4 low
GPT-6 Luna high
Who should choose which
Hy4 Flash with reasoning: the better Tencent choice, for a fraction of Hy4's cost.
Hy4 Flash without reasoning: fast and very cheap ($0.012 per run), for simple work.
Hy4: no setting beat Flash in our suite. Its reasoning runs were also the slowest and most expensive here.
GPT-6 Luna high: a higher score at a similar price, if you do not need a Tencent model.
Availability and prices
OpenRouter describes Hy4 Flash as a cost-efficient Hy4 variant served with roughly a quarter of the flagship's parameter count. Tencent announced Hy4 and Hy4 Flash in October 2026 and has not published Flash's size or weights.
Hy4 Flash has a 1M-token context and takes text and image input. OpenRouter exposes a reasoning on/off switch for it without named effort levels; we tested it with reasoning off and with reasoning on (listed as medium).
Method and limits
Every model ran the same 23 tests three times through OpenRouter. Scores are out of 10, ranks are leaderboard positions at capture time, and comparisons use the 440 complete, non-archived chat results.
Hy4 Flash was tested on November 17, 2026, Hy4 on November 9 and 10, and GPT-6 Luna in September 2026. Tests 14, 15 and 22 kept the same prompts, answers and pass scores; only their recorded evaluator version differs.
The Hy4 Flash price includes OpenRouter's listed 50% endpoint discount, which can end. Response times include provider and workload effects. Costs are recorded spend, not a price guarantee.
Sources
- OpenRouter: Hy4 Flash (accessed 2026-11-17)
- OpenRouter: Hy4 (accessed 2026-11-17)
- TechNode: Tencent says Hy4 and Hy4 Flash will support visual understanding (accessed 2026-11-17)
Linked model pages show current results, which can differ from these figures.